Level set methods for finding critical points of mountain pass type (Q544171)

Let \(X\) be a topological space and \(f: X\to\mathbb{R}\). \(\Gamma(a, b)\) denotes the set of all continuous paths \(p: [0, 1]\to X\) such that \(p(0)= a\) and \(p(1)= b\). A mountain pass \(p^*\in\Gamma(a, b)\) is defined to be a minimizer of the problem \[ \underset{p\in \Gamma(a, b)}{}{\text{inf}}\;\sup_{0\leq t\leq 1}\, f\circ p(t). \] The mountain pass theorem and its variants provide the primary ways to establish the existence of critical points and to find critical points numerically. A numerical method for finding critical points is described in the present paper. This method is convergent in the nonsmooth case and locally superlinearly convergent in the smooth finite-dimensional case. The algorithm allows to find saddle-points of mountain type. These techniques are applied to describe a strategy for adressing the Wilkinson problem of calculating the distance from a matrix to a closest matrix with repeated eigenvalues. The results are illustrated by some examples with convincing figures. There are applications of mountain passes in computational chemistry and in nonlinear partial differential equation.